Working with Ordinal Predictors∗

The use of categorical variables in regression modeling is discussed. Some pitfalls in the use of numerically scaled ordinal variables are considered. ∗Prepared for delivery at the annual meeting of the Midwest Political Science Association, Chicago Illinois, April 2-5, 2009. In the day-to-day practice of social science, nominal and ordinal variables are the pre-dominant variable types. Survey data sets may contain the occasional numerical variable like “age ” (in years), but that type is grossly outweighed by the categorical indicators. Categor-ical variables are usually subdivided into “nominal ” and “ordinal ” types. The use of ordinal variables is the main focus of this paper. The presentation is not aimed at methodologists, but rather at practitioners who work with these very common data types. The basic idea that differentiates ordinal variables is that, in the eye of the person who uses the data, an observation’s membership in one category can somehow be interpreted as “less ” or “more ” than another observation’s membership in a different category. Ordinal variables may be based on judgment scales, so the ordering is not purely in the mind o